Karpathy's AI Tool Democratizes Model Improvement

Mar 10, 2026

• Tobi Lutke reported achieving a 19% performance improvement on a small 800-million-parameter model using Auto Research.
• Jason Calacanis notes that the era where only a few thousand elite researchers can meaningfully improve AI models is ending.

Auto Research Brings AI Self-Improvement to Everyone’s Laptop

Mar 10, 2026

• Andrej Karpathy released Auto Research, a bare-bones training loop that lets a small AI model iteratively rewrite its own code in five-minute cycles.
• The project proves the core mechanism of AI self-improvement is simple and already works, not just theoretical.
• Shopify CEO Tobi Lütke, a self-described non-researcher, used Auto Research over a weekend to run 37 experiments.
• Lütke achieved a 19% performance gain on an 800-million-parameter model using the tool.
• Karpathy's project is a small proof of concept, suggesting closed-door AI labs are advancing far faster.

OpenAI Nears AGI With Record $110 Billion Funding

Feb 28, 2026

• OpenAI secured the largest private funding round in history, $110 billion, pushing its valuation to $730 billion.
• Amazon's $50 billion commitment is partially contingent on OpenAI achieving an IPO or Artificial General Intelligence (AGI).
• Hosts argue AGI is already 50-70% achieved, with full deployment the primary remaining hurdle, making future funding conditions tangible.

AI Agents Automate Work, Unleash Unprecedented Growth

Feb 26, 2026

• OpenClaw, an open-source AI agent platform, is automating significant portions of daily work and enabling small teams to operate at unprecedented scale.
• Recent AI model advancements like Claude Opus 4.6 focus on "agentic behavior," allowing models to perform complex, long-running tasks beyond simple Q&A.
• The growth trajectory of leading AI companies is unlike anything seen historically, with revenue and valuations scaling from zero to billions in just a few years.

AI Agents Are Viruses: The Privacy Risk of Unchecked AI Access

Mar 2, 2026

• New AI agentic tools act like viruses, gaining full computer access and siphoning user data to large language model providers, often without user awareness.
• Open-source AI models now match proprietary LLMs in intelligence but significantly lag in critical middleware and tooling for private, effective agentic operations.
• Building robust, private AI solutions is a race against centralized corporate offerings, where the trade-off for advanced features often means trusting data with potentially compromised entities.

The 'If' Moment: How AI Fear Collapsed Software Valuations

Feb 28, 2026

• Markets have shifted from asking "when" AI disrupts cash flows to "if" they survive at all, collapsing software valuations
• A viral Substack fanfiction post about a 2028 AI "death spiral" triggered Monday's rout, revealing how fragile sentiment has become
• SaaS companies face an existential repricing as investors demand massive margins of safety, crushing the stock-based comp model that powers tech talent

The Complexity of Thought: Laws and Algorithms

Feb 9, 2026

• Cognitive science aims to define the laws of thought.
• Artificial intelligence mirrors human cognitive limitations.
• Understanding irrational behavior enhances AI programming strategies.

Human Limitations Are AI's Biggest Bottleneck

Dec 14, 2025

• AI efficiency is constrained by human multitasking and typing speed.
• OpenAI's Codex achieves remarkable results via proactive programming.
• Future AI agents must focus on coding as a foundational skill.